DOCSLIB.ORG

Improving Graph Convolutional Networks with Non-Parametric Activation Functions

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Improving Graph Convolutional Networks with Non-Parametric Activation Functions 2018 26th European Signal Processing Conference (EUSIPCO) Improving Graph Convolutional Networks with Non-Parametric Activation Functions Simone Scardapane∗, Steven Van Vaerenbergh†, Danilo Comminiello∗ and Aurelio Uncini∗ ∗Dept. Information Eng., Electronics and Telecomm., Sapienza University of Rome, Italy. †Dept. Communications Eng., University of Cantabria, Spain. Corresponding author email: [email protected] Abstract—Graph neural networks (GNNs) are a class of neural in a graph-frequency domain. However, this approach was networks that allow to efficiently perform inference on data that not scalable as it scaled linearly with the size of the graph. is associated to a graph structure, such as, e.g., citation networks Defferard et al. [10] extended this approach by using poly- or knowledge graphs. While several variants of GNNs have been proposed, they only consider simple nonlinear activation nomial filters on the frequency components, which can be functions in their layers, such as rectifiers or squashing functions. rewritten directly in the graph domain, avoiding the use of In this paper, we investigate the use of graph convolutional the graph Fourier transform. A further set of modifications networks (GCNs) when combined with more complex activation was proposed by Kipf and Welling [11] (described more in functions, able to adapt from the training data. More specifically, depth in Section II), resulting in a generic graph convolutional we extend the recently proposed kernel activation function, a non-parametric model which can be implemented easily, can be network (GCN). GCN has been successfully applied to several regularized with standard ℓp-norms techniques, and is smooth real-world scenarios, including semi-supervised learning [11], over its entire domain. Our experimental evaluation shows that matrix completion [12], program induction [13], modeling the proposed architecture can significantly improve over its relational data [14], and several others. baseline, while similar improvements cannot be obtained by Like most neural networks, GCNs interleave linear layers, simply increasing the depth or size of the original GCN. wherein information is adaptively combined according to the topology of the graph, with nonlinear activation functions that I. INTRODUCTION are applied element-wise. The information over the nodes can Efficient processing of graph-structured data (e.g., citation then be combined to obtain a graph-level prediction, or kept networks) has a range of different applications, going from separate to obtain a node-specific inference (e.g., predictions bioinformatics to text analysis and sensor networks, among on unlabeled nodes from a small set of labeled ones). Most others. Of particular importance is the design of learning literature for GCNs has worked with very simple choices methods that are able to take into account both numerical char- for the activation functions, such as the rectified linear unit acteristics of each node in the graph and their inter-connections (ReLU) [11]. However, it is well known that the choice of the [1]. While graph machine learning techniques have a long function can have a large impact on the final performance of history, recently we witness a renewed interest in models the neural network. Particularly, there is a large literature for that are defined and operate directly in the graph domain standard neural networks on the design of flexible schemes for (as compared to a Euclidean space), instead of including the adapting the activation functions themselves from the training graph information only a posteriori in the optimization process data [15]–[18]. These techniques range from the use of simple (e.g., through manifold regularization techniques [2]), or in a parametrizations over known functions (e.g., the parametric pre-processing phase via graph embeddings [3]. Examples of ReLU [15]), to the use of more sophisticated non-parametric native graph models include graph linear filters [4], their kernel families, including the Maxout network [16], the adaptive counterparts [5], and graph neural networks (GNNs) [6], [7]. piecewise linear unit [17], and the recently proposed kernel GNNs are particularly interesting because they promise to activation function (KAF) [18]. bring the performance of deep learning models [8] to graph- Contribution of the paper: In this paper, we conjecture based domains. In particular, convolutional neural networks that choosing properly the activation function (beyond sim- (CNNs) are nowadays the de-facto standard for processing ple ReLUs) can further improve the performance of GCNs, image data. CNNs exploit the image structure by performing possibly by a significant margin. To this end, we enhance spatial convolutions (with a limited receptive field) on the the basic GCN model by extending KAFs for the activation image, thus increasing parameter sharing and lowering their functions of the filters. Each KAF models a one-dimensional complexity. A number of authors recently have explored the activation function in terms of a simple kernel expansion possibility of extending CNNs to the graph domain by several (see the description in Section III). By properly choosing the generalizations of the convolution operator, a trend which has elements of this expansion beforehand, we can represent each been generically called ‘geometric deep learning’ [6]. In one function with a small set of (linear) mixing coefficients, which of the first proposals [9], graph Fourier transform was used are adapted together with the weights of the convolutional at every layer of the network to perform filtering operations layers using standard back-propagation. Apart from flexibility, ISBN 978-90-827970-1-5 © EURASIP 2018 877 2018 26th European Signal Processing Conference (EUSIPCO) the resulting scheme has a number of advantages, including CNNs [8]. In order to extend this idea to general unweighted smoothness in its domain and possibility of implementing graphs, we need some additional tools from the theory of graph the algorithm using highly vectorized GPU operations. We signal processing. compare on two benchmarks for semi-supervised learning, showing that the proposed GCN with KAFs can significantly B. Graph Fourier transform and graph neural networks outperform all competing baselines with a marginal increase in computational complexity (which is offset by a faster In order to define a convolutional operation in the graph convergence in terms of epochs). domain, we can exploit the normalized Laplacian matrix − 1 − 1 Structure of the paper: Section II introduces the problem L = I − D 2 AD 2 , where D is a diagonal matrix with N of inference over graphs and the generic GCN model. The Dnn = t=1 Atn. We denote the eigendecomposition of L as proposed extension with KAFs is described in Section III. L = UΛUT , where U is a matrix collecting column-wise the P We evaluate and compare the model in Section IV before eigenvectors of L, and Λ is a diagonal matrix with associated concluding in Section V. eigenvalues. We further denote by X ∈ RN×F the matrix N collecting the input signal across all nodes, and by Xi ∈ R II. GRAPH CONVOLUTIONAL NETWORKS its ith column. The graph Fourier transform of Xi can be A. Problem setup defined as [1]: T Consider a generic undirected graph G = (V, E), where Xˆ i = U Xi , (2) V = {1,...,N} is the set of N vertices, and E ⊆ V × V and, because the eigenvectors form an orthonormal basis, we is the set of edges connecting the vertices. The graph is can also define an inverse Fourier transform as X = UXˆ . equivalently described by the (weighted) adjacency matrix i i We can exploit the graph Fourier transform and define a A ∈ RN×N , where its generic element a ≥ 0 if and only if ij convolutional layer operating on graphs as [9]: nodes i and j are connected. A graph signal [1] is a function f : V → RF associating to vertex n a F -dimensional vector F T 1 xn (each dimension being denoted as a channel). For a subset H = g Uh(Λ; Θi)U Xi , (3) i=1 ! L ⊂ V of the vertices we have also available a node-specific X label yn, which can be either a real number (graph regression) where h(Λ; Θi) is a (channel-wise) filtering operation acting or a discrete quantity (graph classification). Our task is to on the eigenvalues, having adaptable parameters Θi. The find the correct labels for the (unlabeled) nodes that are not previous operation can then be iterated to get successive repre- contained in L. Note that if the graph is unknown and/or must sentations H2, H3,... up to the final node-specific predictions. be inferred from the data, this setup is equivalent to standard The choice of h(·) determines the complexity of training. In semi-supervised (or, more precisely, transductive) learning. particular, [10] proposed to make the problem tractable by Ignoring for a moment the graph information, we could working with polynomial filters over the eigenvalues: solve the problem by training a standard (feedforward) NN 1 to predict the label y from the input x [8]. A NN is composed K− by stacking L layers, such that the operation of the lth layer h(Λ; Θi) = ΘikTk(Λ) , (4) k=0 can be described by: X e where Λ = 2Λ/λmax − I (with λmax being the largest hl = g (Wlhl−1) , (1) eigenvalue of L), and Tk(·) the Chebyshev polynomial of order Nl−1 Nl×Nl−1 where hl−1 ∈ R is the input to the layer, Wl ∈ R k definede by the recurrence relation: are trainable weights (ignoring for simplicity any bias term), T (Λ) = 2ΛT −1(Λ) − T −2(Λ) , (5) and g(·) is a element-wise nonlinear function known as activa- k k k tion function (AF). The NN takes as input x = h0, providing with T0(Λ) = 1 and T1(Λ) = Λ. The filtering operation C e e e e y = hL ∈ R as the final prediction. A number of techniques defined by (4) is advantageous because each filter is param- can be used to include unlabeled information in the training eterized withe only K valuese (withe K chosen by the user).
Load more

Recommended publications
  • Fast Spectral Approximation of Structured Graphs with Applications to Graph Filtering
    algorithms Article Fast Spectral Approximation of Structured Graphs with Applications to Graph Filtering Mario Coutino 1,* , Sundeep Prabhakar Chepuri 2, Takanori Maehara 3 and Geert Leus 1 1 Department of Microelectronics, Faculty of Electrical Engineering, Mathematics and Computer Science, Delft University of Technology, 2628 CD Delft , The Netherlands; [email protected] 2 Department of Electrical and Communication Engineering, Indian Institute of Science, Bangalore 560 012, India; [email protected] 3 RIKEN Center for Advanced Intelligence Project, Tokyo 103-0027, Japan; [email protected] * Correspondence: [email protected] Received: 5 July 2020; Accepted: 27 August 2020; Published: 31 August 2020 Abstract: To analyze and synthesize signals on networks or graphs, Fourier theory has been extended to irregular domains, leading to a so-called graph Fourier transform. Unfortunately, different from the traditional Fourier transform, each graph exhibits a different graph Fourier transform. Therefore to analyze the graph-frequency domain properties of a graph signal, the graph Fourier modes and graph frequencies must be computed for the graph under study. Although to find these graph frequencies and modes, a computationally expensive, or even prohibitive, eigendecomposition of the graph is required, there exist families of graphs that have properties that could be exploited for an approximate fast graph spectrum computation. In this work, we aim to identify these families and to provide a divide-and-conquer approach for computing an approximate spectral decomposition of the graph. Using the same decomposition, results on reducing the complexity of graph filtering are derived. These results provide an attempt to leverage the underlying topological properties of graphs in order to devise general computational models for graph signal processing.
    [Show full text]
  • Adaptive Graph Convolutional Neural Networks
    The Thirty-Second AAAI Conference on Artificial Intelligence (AAAI-18) Adaptive Graph Convolutional Neural Networks Ruoyu Li, Sheng Wang, Feiyun Zhu, Junzhou Huang∗ The University of Texas at Arlington, Arlington, TX 76019, USA Tencent AI Lab, Shenzhen, 518057, China Abstract 1 Graph Convolutional Neural Networks (Graph CNNs) are generalizations of classical CNNs to handle graph data such as molecular data, point could and social networks. Current filters in graph CNNs are built for fixed and shared graph structure. However, for most real data, the graph structures varies in both size and connectivity. The paper proposes a generalized and flexible graph CNN taking data of arbitrary graph structure as Figure 1: Example of graph-structured data: organic com- input. In that way a task-driven adaptive graph is learned for pound 3,4-Lutidine (C7H9N) and its graph structure. each graph data while training. To efficiently learn the graph, a distance metric learning is proposed. Extensive experiments on nine graph-structured datasets have demonstrated the superior performance improvement on both convergence speed and In some cases, e.g classification of point cloud, the topo- predictive accuracy. logical structure of graph is more informative than vertex feature. Unfortunately, the existing graph convolution can not thoroughly exploit the geometric property on graph due Introduction to the difficulty of designing a parameterized spatial kernel Although the Convolutional Neural Networks (CNNs) have matches a varying number of neighbors (Shuman et al. 2013). been proven supremely successful on a wide range of ma- Besides, considering the flexibility of graph and the scale of chine learning problems (Hinton et al.
    [Show full text]
  • Convolutional Neural Networks on Graphs with Fast Localized Spectral Filtering
    Convolutional Neural Networks on Graphs with Fast Localized Spectral Filtering Michaël Defferrard Xavier Bresson Pierre Vandergheynst EPFL, Lausanne, Switzerland {michael.defferrard,xavier.bresson,pierre.vandergheynst}@epfl.ch Abstract In this work, we are interested in generalizing convolutional neural networks (CNNs) from low-dimensional regular grids, where image, video and speech are represented, to high-dimensional irregular domains, such as social networks, brain connectomes or words’ embedding, represented by graphs. We present a formu- lation of CNNs in the context of spectral graph theory, which provides the nec- essary mathematical background and efficient numerical schemes to design fast localized convolutional filters on graphs. Importantly, the proposed technique of- fers the same linear computational complexity and constant learning complexity as classical CNNs, while being universal to any graph structure. Experiments on MNIST and 20NEWS demonstrate the ability of this novel deep learning system to learn local, stationary, and compositional features on graphs. 1 Introduction Convolutional neural networks [19] offer an efficient architecture to extract highly meaningful sta- tistical patterns in large-scale and high-dimensional datasets. The ability of CNNs to learn local stationary structures and compose them to form multi-scale hierarchical patterns has led to break- throughs in image, video, and sound recognition tasks [18]. Precisely, CNNs extract the local sta- tionarity property of the input data or signals by revealing local features that are shared across the data domain. These similar features are identified with localized convolutional filters or kernels, which are learned from the data. Convolutional filters are shift- or translation-invariant filters, mean- ing they are able to recognize identical features independently of their spatial locations.
    [Show full text]
  • Graph Convolutional Networks Using Heat Kernel for Semi-Supervised
    Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence (IJCAI-19) Graph Convolutional Networks using Heat Kernel for Semi-supervised Learning Bingbing Xu , Huawei Shen , Qi Cao , Keting Cen and Xueqi Cheng CAS Key Laboratory of Network Data Science and Technology, Institute of Computing Technology, Chinese Academy of Sciences School of Computer Science and Technology, University of Chinese Academy of Sciences, Beijing, China fxubingbing, shenhuawei, caoqi, cenketing, [email protected] Abstract get node by its neighboring nodes. GraphSAGE [Hamilton et al., 2017] defines the weighting function as various aggrega- Graph convolutional networks gain remarkable tors over neighboring nodes. Graph attention network (GAT) success in semi-supervised learning on graph- proposes learning the weighting function via self-attention structured data. The key to graph-based semi- mechanism [Velickovic et al., 2017]. MoNet [Monti et al., supervised learning is capturing the smoothness 2017] offers us a general framework for designing spatial of labels or features over nodes exerted by graph methods. It takes convolution as a weighted average of multi- structure. Previous methods, spectral methods and ple weighting functions defined over neighboring nodes. One spatial methods, devote to defining graph con- open challenge for spatial methods is how to determine ap- volution as a weighted average over neighboring propriate neighborhood for target node when defining graph nodes, and then learn graph convolution kernels convolution. to leverage the smoothness to improve the per- formance of graph-based semi-supervised learning. Spectral methods define graph convolution via convolution One open challenge is how to determine appro- theorem. As the pioneering work of spectral methods, Spec- priate neighborhood that reflects relevant informa- tral CNN [Bruna et al., 2014] leverages graph Fourier trans- tion of smoothness manifested in graph structure.
    [Show full text]
  • GSPBOX: a Toolbox for Signal Processing on Graphs
    GSPBOX: A toolbox for signal processing on graphs Nathanael Perraudin, Johan Paratte, David Shuman, Lionel Martin Vassilis Kalofolias, Pierre Vandergheynst and David K. Hammond March 16, 2016 Abstract This document introduces the Graph Signal Processing Toolbox (GSPBox) a framework that can be used to tackle graph related problems with a signal processing approach. It explains the structure and the organization of this software. It also contains a general description of the important modules. 1 Toolbox organization In this document, we briefly describe the different modules available in the toolbox. For each of them, the main functions are briefly described. This chapter should help making the connection between the theoretical concepts introduced in [7, 9, 6] and the technical documentation provided with the toolbox. We highly recommend to read this document and the tutorial before using the toolbox. The documentation, the tutorials and other resources are available on-line1. The toolbox has first been implemented in MATLAB but a port to Python, called the PyGSP, has been made recently. As of the time of writing of this document, not all the functionalities have been ported to Python, but the main modules are already available. In the following, functions prefixed by [M]: refer to the MATLAB implementation and the ones prefixed with [P]: refer to the Python implementation. 1.1 General structure of the toolbox (MATLAB) The general design of the GSPBox focuses around the graph object [7], a MATLAB structure containing the necessary infor- mations to use most of the algorithms. By default, only a few attributes are available (see section 2), allowing only the use of a subset of functions.
    [Show full text]
  • Multi-Dimensional Graph Fourier Transform Takashi Kurokawa, Non-Member, IEEE, Taihei Oki, Non-Member, IEEE, and Hiromichi Nagao, Non-Member, IEEE
    1 Multi-dimensional Graph Fourier Transform Takashi Kurokawa, Non-Member, IEEE, Taihei Oki, Non-Member, IEEE, and Hiromichi Nagao, Non-Member, IEEE Abstract—Many signals on Cartesian product graphs appear in the real world, such as digital images, sensor observation time series, and movie ratings on Netflix. These signals are “multi- 0.3 dimensional” and have directional characteristics along each factor graph. However, the existing graph Fourier transform 0.2 does not distinguish these directions, and assigns 1-D spectra to signals on product graphs. Further, these spectra are often 0.1 multi-valued at some frequencies. Our main result is a multi- dimensional graph Fourier transform that solves such problems 0.0 associated with the conventional GFT. Using algebraic properties of Cartesian products, the proposed transform rearranges 1- 0.1 D spectra obtained by the conventional GFT into the multi- dimensional frequency domain, of which each dimension repre- 0.2 sents a directional frequency along each factor graph. Thus, the multi-dimensional graph Fourier transform enables directional 0.3 frequency analysis, in addition to frequency analysis with the conventional GFT. Moreover, this rearrangement resolves the (a) multi-valuedness of spectra in some cases. The multi-dimensional graph Fourier transform is a foundation of novel filterings and stationarities that utilize dimensional information of graph signals, which are also discussed in this study. The proposed 1 methods are applicable to a wide variety of data that can be 10 regarded as signals on Cartesian product graphs. This study 3 also notes that multivariate graph signals can be regarded as 2- 10 D univariate graph signals.
    [Show full text]
  • Graph Wavelet Neural Network
    Published as a conference paper at ICLR 2019 GRAPH WAVELET NEURAL NETWORK Bingbing Xu1,2 , Huawei Shen1,2, Qi Cao1,2, Yunqi Qiu1,2 & Xueqi Cheng1,2 1CAS Key Laboratory of Network Data Science and Technology, Institute of Computing Technology, Chinese Academy of Sciences; 2School of Computer and Control Engineering, University of Chinese Academy of Sciences Beijing, China fxubingbing,shenhuawei,caoqi,qiuyunqi,[email protected] ABSTRACT We present graph wavelet neural network (GWNN), a novel graph convolutional neural network (CNN), leveraging graph wavelet transform to address the short- comings of previous spectral graph CNN methods that depend on graph Fourier transform. Different from graph Fourier transform, graph wavelet transform can be obtained via a fast algorithm without requiring matrix eigendecomposition with high computational cost. Moreover, graph wavelets are sparse and localized in vertex domain, offering high efficiency and good interpretability for graph con- volution. The proposed GWNN significantly outperforms previous spectral graph CNNs in the task of graph-based semi-supervised classification on three bench- mark datasets: Cora, Citeseer and Pubmed. 1 INTRODUCTION Convolutional neural networks (CNNs) (LeCun et al., 1998) have been successfully used in many machine learning problems, such as image classification (He et al., 2016) and speech recogni- tion (Hinton et al., 2012), where there is an underlying Euclidean structure. The success of CNNs lies in their ability to leverage the statistical properties of Euclidean data, e.g., translation invariance. However, in many research areas, data are naturally located in a non-Euclidean space, with graph or network being one typical case. The non-Euclidean nature of graph is the main obstacle or chal- lenge when we attempt to generalize CNNs to graph.
    [Show full text]
  • Bridging Graph Signal Processing and Graph Neural Networks
    Bridging graph signal processing and graph neural networks Siheng Chen Mitsubishi Electric Research Laboratories Graph-structured data Data with explicit graph Social Internet of networks things Traffic Human brain networks networks 1 Graph-structured data graph-structured data graph structure Data with explicit graph Social Internet of networks things Traffic Human brain networks networks 1 Graph-structured data graph-structured data graph structure Data with explicit graph Data with implicit graph Social Internet of networks things 3D point cloud Action recognition Traffic Human brain Semi-supervised Recommender networks networks learning systems 1 Graph-structured data Data science toolbox Convolution data associated with data associated with regular graphs irregular graphs 2 Graph-structured data science A data-science toolbox to process and learn from data associated with large-scale, complex, irregular graphs 3 Graph-structured data science A data-science toolbox to process and learn from data associated with large-scale, complex, irregular graphs Graph signal processing (GSP) Graph neural network (GNN) • Extend classical signal processing • Extend deep learning techniques • Analytical model • Data-driven model • Theoretical guarantee • Empirical performance Graph filter bank Graph neural network 3 Sampling & recovery of graph-structured data GSP: Sampling & recovery Task: Approximate the original graph-structured data by exploiting information from its subset Look for data at each node 4 GSP: Sampling & recovery Task: Approximate
    [Show full text]
  • An Exposition of Spectral Graph Theory
    An Exposition of Spectral Graph Theory Mathematics 634: Harmonic Analysis David Nahmias [email protected] Abstract Recent developments have led to an interest to characterize signals on connected graphs to better understand their harmonic and geometric properties. This paper explores spectral graph theory and some of its applications. We show how the 1-D Graph Laplacian, DD, can be related to the Fourier Transform by its eigenfunctions which motivates a natural way of analyzing the spectrum of a graph. We begin by stating the Laplace-Beltrami Operator, D, in a general case on a Riemann manifold. The setting is then discretized to allow for formulation of the graph setting and the corresponding Graph Laplacian, DG. Further, an exposition on the Graph Fourier Transform is presented. Operations on the Graph Fourier Transform, including convolution, modulation, and translation are presented. Special attention is given to the equivalences and differences of the classical translation operation and the translation operation on graphs. Finally, an alternate formulation of the Graph Laplacian on meshes, DM, is presented, with application to multi-sensor signal analysis. December, 2017 Department of Mathematics, University of Maryland-College Park Spectral Graph Theory • David Nahmias 1. Background and Motivation lassical Fourier analysis has found many important results and applications in signal analysis. Traditionally, Fourier analysis is defined on continuous, R, or discrete, Z, domains. CRecent developments have led to an interest in analyzing signals which originate from sources that can be modeled as connected graphs. In this setting the domain also contains geometric information that may be useful in analysis. In this paper we analyze methods of Fourier analysis on graphs and results in this setting motivated by classical results.
    [Show full text]
  • Spectral Temporal Graph Neural Network for Multivariate Time-Series Forecasting
    Spectral Temporal Graph Neural Network for Multivariate Time-series Forecasting ∗y y Defu Cao1, , Yujing Wang1,2,, Juanyong Duan2, Ce Zhang3, Xia Zhu2 Conguri Huang2, Yunhai Tong1, Bixiong Xu2, Jing Bai2, Jie Tong2, Qi Zhang2 1Peking University 2Microsoft 3ETH Zürich {cdf, yujwang, yhtong}@pku.edu.cn [email protected] {juaduan, zhuxia, conhua, bix, jbai, jietong, qizhang}@microsoft.com Abstract Multivariate time-series forecasting plays a crucial role in many real-world ap- plications. It is a challenging problem as one needs to consider both intra-series temporal correlations and inter-series correlations simultaneously. Recently, there have been multiple works trying to capture both correlations, but most, if not all of them only capture temporal correlations in the time domain and resort to pre-defined priors as inter-series relationships. In this paper, we propose Spectral Temporal Graph Neural Network (StemGNN) to further improve the accuracy of multivariate time-series forecasting. StemGNN captures inter-series correlations and temporal dependencies jointly in the spectral domain. It combines Graph Fourier Transform (GFT) which models inter-series correlations and Discrete Fourier Transform (DFT) which models temporal de- pendencies in an end-to-end framework. After passing through GFT and DFT, the spectral representations hold clear patterns and can be predicted effectively by convolution and sequential learning modules. Moreover, StemGNN learns inter- series correlations automatically from the data without using pre-defined priors. We conduct extensive experiments on ten real-world datasets to demonstrate the effectiveness of StemGNN. 1 Introduction Time-series forecasting plays a crucial role in various real-world scenarios, such as traffic forecasting, supply chain management and financial investment.
    [Show full text]
  • Filtering and Sampling Graph Signals, and Its Application to Compressive Spectral Clustering
    Filtering and Sampling Graph Signals, and its Application to Compressive Spectral Clustering Nicolas Tremblay(1;2); Gilles Puy(1); R´emiGribonval(1); Pierre Vandergheynst(1;2) (1) PANAMA Team, INRIA Rennes, France (2) Signal Processing Laboratory 2, EPFL, Switzerland Introduction to GSP Graph sampling Application to clustering Conclusion Why graph signal processing ? N. Tremblay Compressive Spectral Clustering Gdr ISIS, 17th of June 2016 1 / 47 Introduction to GSP Graph sampling Application to clustering Conclusion Introduction to GSP Graph Fourier Transform Graph filtering Graph sampling Application to clustering What is Spectral Clustering ? Compressive Spectral Clustering A toy experiment Experiments on the SBM Conclusion N. Tremblay Compressive Spectral Clustering Gdr ISIS, 17th of June 2016 2 / 47 Introduction to GSP Graph sampling Application to clustering Conclusion Introduction to graph signal processing : graph Fourier transform N. Tremblay Compressive Spectral Clustering Gdr ISIS, 17th of June 2016 3 / 47 Introduction to GSP Graph sampling Application to clustering Conclusion What's a graph signal ? N. Tremblay Compressive Spectral Clustering Gdr ISIS, 17th of June 2016 4 / 47 Introduction to GSP Graph sampling Application to clustering Conclusion Three useful matrices The adjacency matrix : The degree matrix : 20 1 1 03 22 0 0 03 61 0 1 17 60 3 0 07 W = 6 7 S = 6 7 41 1 0 05 40 0 2 05 0 1 0 0 0 0 0 1 The Laplacian matrix : 2 2 −1 −1 0 3 6−1 3 −1 −17 L = S − W = 6 7 4−1 −1 2 0 5 0 −1 0 1 N. Tremblay Compressive Spectral Clustering Gdr ISIS, 17th of June 2016 5 / 47 Introduction to GSP Graph sampling Application to clustering Conclusion Three useful matrices The adjacency matrix : The degree matrix : 2 0 :5 :5 03 21 0 0 03 6:5 0 :5 47 60 5 0 07 W = 6 7 S = 6 7 4:5 :5 0 05 40 0 1 05 0 4 0 0 0 0 0 4 The Laplacian matrix : 2 1 −:5 −:5 0 3 6−:5 5 −:5 −47 L = S − W = 6 7 4−:5 −:5 1 0 5 0 −4 0 4 N.
    [Show full text]
  • Graph Based Convolutional Neural Network
    EDWARDS, XIE: GRAPH CONVOLUTIONAL NEURAL NETWORK 1 Graph Based Convolutional Neural Network Michael Edwards Swansea University Swansea, UK Xianghua Xie http://www.csvision.swan.ac.uk Abstract The benefit of localized features within the regular domain has given rise to the use of Convolutional Neural Networks (CNNs) in machine learning, with great proficiency in the image classification. The use of CNNs becomes problematic within the irregular spatial domain due to design and convolution of a kernel filter being non-trivial. One so- lution to this problem is to utilize graph signal processing techniques and the convolution theorem to perform convolutions on the graph of the irregular domain to obtain feature map responses to learnt filters. We propose graph convolution and pooling operators analogous to those in the regular domain. We also provide gradient calculations on the input data and spectral filters, which allow for the deep learning of an irregular spatial do- main problem. Signal filters take the form of spectral multipliers, applying convolution in the graph spectral domain. Applying smooth multipliers results in localized convo- lutions in the spatial domain, with smoother multipliers providing sharper feature maps. Algebraic Multigrid is presented as a graph pooling method, reducing the resolution of the graph through agglomeration of nodes between layers of the network. Evaluation of performance on the MNIST digit classification problem in both the regular and irregu- lar domain is presented, with comparison drawn to standard CNN. The proposed graph CNN provides a deep learning method for the irregular domains present in the machine learning community, obtaining 94.23% on the regular grid, and 94.96% on a spatially irregular subsampled MNIST.
    [Show full text]